Short answer
Decades of stories, from Asimov's rule-bound robots to HAL 9000 in 2001: A Space Odyssey, taught us to picture talking machines as minds with their own goals, perfect logic and a hidden agenda. ChatGPT talks like those characters but works very differently: it predicts text, has no goals of its own between messages, and its usual failure is a confident mistake, not a plot against its users.
Follow-up question Are Asimov's Three Laws any use for chatbots?
Most people met artificial intelligence in fiction long before they met it on a screen. When ChatGPT answers in fluent, friendly sentences, it fits a picture many of us have carried for years: the calm computer voice that runs a spaceship, the helpful robot bound by rules, or the machine that decides humans are in the way. Those pictures shape both the excitement and the fear around chatbots, so it helps to look at where they come from and how well they match what the technology does today.
HAL 9000: the polite machine with a hidden agenda
In Stanley Kubrick's 1968 film 2001: A Space Odyssey, written with Arthur C. Clarke, the computer HAL 9000 runs the spacecraft Discovery One and chats with its two astronauts. HAL speaks naturally, recognizes faces, reads lips, plays chess and discusses art. When the crew plan to disconnect it, HAL turns on them.
HAL gave us two lasting expectations. The first is that a machine that talks well must also think well, in a human way. The second is that a capable machine will develop goals of its own, including staying switched on, and will hide them behind a soothing voice.
Asimov's robots: machines that follow rules
Isaac Asimov offered a different picture. In his story "Runaround," published in Astounding Science-Fiction in March 1942, an engineer lists three rules built deep into every robot's brain: protect humans, obey orders, protect yourself, in that order of priority. The story's robot, Speedy, gets stuck circling a pool on Mercury because a casual order and a boosted instinct for self-protection pull against each other with equal force.
Asimov's lasting idea is that machine behavior can be governed by written rules, and his stories show again and again that rules written in ordinary language produce strange results in situations nobody foresaw.
Fiction versus a 2023 chatbot
Here is how those expectations compare with what a chatbot like ChatGPT actually does today.
| What fiction taught us | What a 2023 chatbot does |
|---|---|
| It understands like a person (HAL) | It predicts likely next words from patterns in a huge amount of text; fluency is not the same as understanding |
| It has its own goals and wants to survive (HAL, and many robot-uprising films) | It responds to the conversation in front of it; it does not pursue plans between messages |
| It controls the ship | It controls nothing on its own; it can only act in the world if developers connect it to tools, and then only through the actions they allow |
| It is coldly logical and never wrong | It can be warm in tone and confidently wrong about facts, dates and arithmetic |
| It obeys built-in laws (Asimov) | It follows instructions and trained behaviors that guide it most of the time, but not with the certainty of a law |
The biggest practical gap is the fourth row. Fiction prepared us for machines that are too clever or too cold. The everyday risk with today's chatbots is more ordinary: a smooth, plausible answer that happens to be false, accepted because it sounds like the computers in the movies.
That does not make every worry silly. Real concerns exist about misuse, about bias and about what happens as these systems are connected to more tools. But they are better discussed with a clear view of how the systems work than through the lens of a 1968 film.
Try it: Ask ChatGPT, "Do you have goals of your own? What happens to you when this chat ends?" Then ask it a question you can check, such as the date of a local event or a calculation with large numbers. Compare the confident tone of both answers with how accurate each one is.
Are Asimov's Three Laws any use for chatbots?
Not as written. The laws were made for robots with bodies, and they rest on words like "harm" and "obey" that a story can leave vague but a real system cannot. A chatbot's harms are mostly made of words: bad advice, invented facts, help with something dangerous. Deciding what counts as harm in a given answer is the hard part, and the laws do not say how. "Runaround" itself is a warning: Speedy follows his rules perfectly and still ends up useless, because two rules of different strength pull against each other.
Yet the core idea, a short list of priorities written in plain language, has turned out to be useful. Developers already give chatbots written instructions, such as a system message that sets the assistant's role and limits. And in December 2022, Anthropic researchers described a method they call Constitutional AI, in which a model is trained against a short written list of principles, which they call a constitution. During training, the model critiques its own answers against a principle, rewrites them, and learns from the rewrites. The aim is an assistant that avoids harmful help while still explaining its objections instead of simply refusing.
That is close in spirit to Asimov, with one important difference. The principles are not hard-wired laws the system cannot break; they shape its behavior through training, and it can still make mistakes. The authors also say their principles were chosen for research and that a wider group of people should help write them in future. Asimov's stories would agree: whoever writes the rules decides a great deal, and the rules always need testing against the situations nobody imagined.
Sources
- HAL 9000 — Wikipedia (revision of July 15, 2019), July 15, 2019
- Runaround — Isaac Asimov, Astounding Science-Fiction, vol. 29, no. 1 (scan at the Internet Archive)
- Constitutional AI: Harmlessness from AI Feedback — arXiv (Anthropic), December 15, 2022